What Men Want: Politicians’ Strategic Response to Gender Quotas

Julia Michal Clark, Alexandra Domike Blackman, Aytug˘ S¸as¸maz* Word Count: 11870 *DRAFT*

April 12, 2021

Abstract

Tunisia’s 2018 municipal elections, the first since the 2011 revolution, included the adop- tion of strict gender quotas that resulted in near-parity of male and female elected councillors. Despite this massive achievement for descriptive representation, fewer than 20 percent of the mayors – selected from among these councillors – were women. We argue that this gender gap in council leadership is the result of parties’ strategic behavior. To avoid “displacing” male leaders, parties placed female-headed lists (FHLs) in less important municipalities and those in which their previous electoral performance was weakest. By concentrating FHLs in weaker municipalities, female candidates were at a disadvantage during the mayoral selection process. We provide evidence of this theory using election data, an original survey of municipal council candidates, and interviews. This research highlights the role that party elites play in maintain- ing the existing political bargain at the expense of underrepresented groups, even where strict quotas are adopted.

*We would like to thank the U.S. Middle East Partnership Initiative (MEPI), Democracy International, the Freeman Spogli Institute (Stanford University), the Abbasi Program in Islamic Studies (Stanford), the Hicham Alaoui Foundation and the Program on Governance & Local Development (Gothenburg University), the Institute for Quantitative Social Science (Harvard University), the Center for Middle Eastern Studies (Harvard University), and the Project on Middle East Political Science for their generous support of this project. We thank Mariel Barnes, Carolyn Barnett, Lindsay Benstead, Killian Clarke, Michael Hoffman, Christiana Parreira, Daniel Tavana, Nicolas van de Walle, and Michelle Weitzel for detailed comments on earlier drafts. We are also grateful for the helpful feedback on earlier drafts from seminar participants in the Gender in MENA Politics Research Group, at NYU-Abu Dhabi, at Cornell, and at the 2019 APSA conference. We thank ELKA Consulting in Tunis for administering the candidate survey. Intissar Samarat, Manaa Lajnef, Khaled Ferjani, Karim Boudlal, and Houda Ould Khal provided excellent research assistance. All errors and omissions remain ours. 1 Introduction Why do women continue to be underrepresented in politics and leadership positions, despite the adoption of formal institutions like quotas that—at least on paper—guarantee parity? Evidence from a number of countries suggests that the gap between mandated representation and the ac- tual number of women elected is a function of the strength of the quota and other electoral rules (Htun and Jones, 2002; Jones, 2009) and the persistence of informal institutions and norms that shape candidate recruitment and electoral strategies. In particular, the literature has highlighted concretely how male-dominated party leadership and networks undermine the impact of quotas and the selection of women for leadership positions (Esteve-Volart and Bagues, 2012; Lassebie,´ 2020).

Tunisia’s 2018 municipal elections provide a unique setting to better understand these dynamics, including how changes in formal rules for representation interact with parties’ recruitment and electoral strategies. Although Tunisia previously adopted quotas for national-level elections, law- makers strengthened these measures to guarantee that women would be 50 percent of candidates and heads of electoral lists for its 2018 municipal elections. As a result of these quotas, 47 percent of elected municipal councillors were women, a record achievement for women in Tunisia that is matched by few countries in the world. However, when it came time to indirectly elect may- ors from among the elected municipal councillors, fewer than 20 percent of municipalities chose a woman.1 Therefore, while quotas helped close the descriptive representation gap in Tunisia’s council membership, they have not fully equalized power for women within local government, particularly at the highest levels.

In this article, we demonstrate that the relative under-representation of women in municipal coun- cil leadership is largely the result of parties’ strategic placement of female-headed lists (FHLs) in less important municipalities and those in which the parties’ previous electoral performance was weakest. By concentrating FHLs in weaker municipalities, female candidates were at a disad- vantage during the mayoral selection process. Moreover, even in municipalities where FHLs did receive the highest share of votes, women were selected as mayors at lower rates. In contrast to theories that these strategies are primarily motivated by innate bias against women or a lack of qualified female candidates, we argue that parties’ behavior is better explained by the dominance of men in existing political leadership and recruitment networks, which acted to maintain the status quo.

This article makes two main contributions to the study of institutions, women’s representation, and local politics. First, we add to the growing literature on the effect of gender quotas on women’s representation and party strategies (e.g., Chattopadhyay and Duflo, 2004; Frechette´ et al., 2008;

1The council presidents, or mayors, are indirectly elected by the council members after the election. Only the heads of electoral lists are eligible to be mayor unless there is a resignation.

1 Beaman et al., 2009; Hughes, 2011; Esteve-Volart and Bagues, 2012; Lassebie,´ 2020) and extend this to the North African region, where empirical work on this topic has been more limited.2 Second, we use a novel survey of candidates for the local election to add depth to our understanding of the reasons why parties attempt to manipulate quota rules, and demonstrate that this goes beyond concerns about voter bias or candidate qualifications.

These findings have implications not only for women’s representation in Tunisia, but our broader understanding of electoral strategies adopted in response to measures intended to broaden par- ticipation in governance. Institutional engineering may be a first step to increasing the number of women or other historically marginalized groups in political office, but it has its limits when con- fronted with strong motivation for non-compliance. Furthermore, even where quotas and other institutional reforms do succeed in achieving equal representation in numbers, they may be insuf- ficient to ensure an equal share of power, at least in the short run.

We begin by establishing theoretical expectations for how parties behave strategically under quo- tas, and then present the details of the Tunisian case and 2018 municipal elections. Following an overview of our specific hypotheses, data, and methods, we present our results and explore possi- ble explanations for our findings. We conclude by discussing the implications of parties’ strategic behavior for women’s representation and political development in Tunisia and beyond.

2 Gender Quotas and Party Strategy Despite significant progress over the past decades, women’s representation in elected office lags behind men in most countries. Globally, women make up only 25 percent of representatives elected to national legislatures, and only 21 percent of cabinet ministers (World Economic Forum, 2020). Recent work has identified several obstacles to increasing the presence of women in political of- fice, including voter bias against female candidates, gendered differences in political ambition and professional networks, and fewer women in the political “pipeline” (Lawless and Fox, 2005; Crowder-Meyer, 2013; Teele et al., 2018; Blackman and Jackson, 2019; Thomsen and King, 2020).

In recognition of these barriers and of the value of improving both descriptive and substantive rep- resentation (Mansbridge, 1999), policy proposals have focused on addressing pipeline issues by encouraging women to run for lower offices earlier and implementing special training programs for women, as well as adopting gender quotas that mandate a certain level of female candidates or elected officials. However, evidence suggests that even with such measures in place, the gap in po- litical representation persists, often as a result of gatekeepers’ decisions to place female candidates in weaker districts (e.g., Thomas and Bodet, 2013) and male-dominated recruitment methods and

2For notable exceptions, see: Belschner (2018); Belschner and Garcia de Paredes (2020); Shalaby and Elimam (2020). However, in contrast to our focus on the placement of FHLs, these studies focus primarily on how gender and youth quotas interact in terms of the types of candidates that parties select overall, and on the factors driving the adoption of such quotas following the Arab Spring.

2 networks (e.g., Crowder-Meyer, 2013).

Our study builds on this research. By examining a case where strict gender quotas have ensured gender parity in the number of female candidates and significantly narrowed the scope of po- tential strategic behavior, we can better examine political parties’ roles in maintaining the gender gap even in a best-case scenario. This section provides an overview of quota design and theo- retical expectations for parties’ behavior in response to these measures, as well as their potential motivations for circumventing the spirit—if not the letter—of quota laws.

2.1 Quotas Rules and Effectiveness Gender quotas have become a cornerstone of the political process in a majority of the world’s democracies. As of 2020, over 80 countries have implemented some form of legislated gender quota in elections for their national elections, while over 70 have adopted legislated quotas for lower levels of government, including state or provincial governments as well as municipalities and other local councils (International IDEA, 2020).

While some quota systems reserve a specific number of seats within an legislative body (seat reser- vations), the most common type of system seeks to increase the share of female candidates running in an election (candidate quotas) (International IDEA, 2020). In proportional representation (PR) systems or others with multi-member districts (MMDs), candidate quotas typically take the form of a specified percent of candidates within a given electoral list or set of lists that must be women. In single-member districts (SMDs) systems, they generally mandate a percent of a party’s total candidates across all districts that must be women. In addition, some quotas also include “place- ment mandates” that specify requirements for the position of a female candidate within a list (for example, alternating male and female candidates, or requiring a specific number of women at the top of each list) (Htun and Jones, 2002).

Countries with quotas typically have a higher proportion of female representatives than those without quotas (e.g., Jones, 2004; Schmidt and Saunders, 2004). Furthermore, a growing body of work suggests that quotas can have broad positive impacts including improving women’s engage- ment in politics, increasing investment in female-preferred public goods, reducing corruption, and improving the quality of policymaking and administration more broadly (e.g., Chattopadhyay and Duflo, 2004; Barnes and Burchard, 2012; De Paola et al., 2014; Braga and Scervini, 2017; Besley et al., 2017; Lindgren et al., 2009; Benstead, 2019).3

However, the potentially positive substantive effects of gender quotas rest on their ability to actu- ally increase the number of women elected and, importantly, the number who reach positions of power. As a number of studies have shown, the effectiveness of quota policies varies significantly by their design and the type of electoral system in which they are embedded. In particular, global

3However, the positive impact of quotas on substantive outcomes is not guaranteed. Clayton (2015), for example, finds that quotas in Lesotho had a null or negative effect on women’s reported engagement with politics.

3 evidence suggests suggests that the most effective quotas are those in closed-list PR systems that are legislated (instead of voluntary), have strict penalties that are enforced, and have placement mandates (Htun and Jones, 2002; Jones, 2009; Schwindt-Bayer, 2009; Pande and Ford, 2012).

Yet even with relatively strong quotas in place, there is often a loss of fidelity between the man- dated proportion of female candidates and the actual number of women elected to office or leader- ship positions. For example, in French municipalities that require 50 percent female candidates in alternating list positions (so-called “zippered lists”), only 17 percent of larger municipalities and 13 percent of smaller municipalities had a female mayor in 2014 (Lassebie,´ 2020). These inequali- ties are often the direct result of parties’ strategic actions to circumvent or manipulate quota rules to reduce the cost of compliance (Frechette´ et al., 2008; Esteve-Volart and Bagues, 2012).

2.2 Party Strategies in Response to Quotas By design, quotas are intended to increase the number of women in elected office beyond that achieved under the status quo. Quotas will therefore impose costs to the extent that parties4 must adjust their behavior to recruit or run more women than before.5 In extreme cases, parties may be able to avoid quotas altogether by choosing not to comply if the penalty for doing so is unenforced or less than the cost of changing their behavior (Jones, 2009). In France, for example, many parties chose to pay a fee rather than adhere to the quota (Frechette´ et al., 2008). Where non-compliance is not a viable option—e.g., the penalty for non-compliance is disqualification—parties are likely to use other strategies to reduce the impact of the quotas.

The differential placement of female candidates within lists or across districts is one of the most common strategies that parties employ to this effect.6 This can take many forms, but the over- all effect is that women are typically placed where they are more likely to lose, while winning positions are reserved for men. For example, in closed-list PR systems and others where parties determine list rank, parties have often placed female candidates at the bottom of the list and re- served list-head seats for men (Schwindt-Bayer, 2009; Esteve-Volart and Bagues, 2012; Franceschet and Piscopo, 2012). In SMDs and other systems, female candidates have also often been placed in districts where the party expects to lose, serving as “sacrificial lambs” with little hope of winning (Thomas and Bodet, 2013).

Countries have adopted measures to try and limit this type of manipulation, including stricter penalties for non-compliance and adopting placement mandates to avoid pooling female candi- dates at the bottom of the list. However, parties have found creative ways to subvert these man-

4We use the term “parties” here and throughout the article as a shorthand to refer to party selectors or gatekeepers involved in the candidate recruitment process, as described more fully in Norris and Lovenduski (1995). 5In this article, we focus on strategies for complying with quotas, conditional on their adoption. However, parties may also be strategic in their decision to adopt quotas (Frechette´ et al., 2008; Weeks, 2018). 6Depending on the systems, parties may have other strategies available to them, including forming breakaway parties or lists, or running as independents (Lassebie,´ 2020).

4 dates. For example, Esteve-Volart and Bagues (2012) find that in Spanish Senate elections where list order is determined alphabetically by surname, parties ensure that male candidates have the most favorable ballot positions by choosing female candidates with surnames that are lower in the alphabet. Thus, although stronger rules reduce the number and type of available strategies, there may be limits to institutional engineering; with the will to lessen the impact of quotas, parties will often find a way.

2.3 Motivations for Quota Manipulation Parties may seek to reduce their compliance with the quotas for multiple, interrelated reasons that affect the precise strategies they adopt. Overall, the existing literature focuses on three potential drivers of this behavior: (1) voter or party bias against female candidates, (2) a lack of qualified female candidates, and (3) a desire to prevent the displacement of male leaders.

The first is a “demand-side” explanation: parties prefer male candidates because they are con- cerned about voter bias against women and/or are biased against women themselves (Norris and Loven- duski, 1995; Pande and Ford, 2012). There is reason to assume that voter or party bias plays a role in selecting and placing female candidates, particularly in societies with entrenched patriarchal norms and gender roles and where politics have historically been dominated by men. If voters— or party leaders—hold attitudes against women serving in leadership roles or perceive women to be less qualified for office, parties are likely to prefer male candidates. Such biases against women in politics are well-documented in a variety of contexts,7 including in Tunisia (Benstead et al., 2015; Blackman and Jackson, 2019), and multiple studies have demonstrated penalties for female candidates at the ballot box (Kelley and McAllister, 1984; Frechette´ et al., 2008).

The second potential motivation to manipulate quotas is a “supply-side” argument: parties may have difficulty finding qualified female candidates who are willing to run for office on their ticket. Indeed, a commonly voiced concern in societies that have newly adopted quotas is that “quota candidates” will be less qualified than their male counterparts (Jones, 2004; Franceschet and Pis- copo, 2012; Murray, 2014). Even if women have equal capacity for leadership, they may be less likely to have the requisite qualifications (e.g., holding advanced degrees or previous leadership experience), particularly in contexts where fewer women have been involved in the public sphere (Pande and Ford, 2012).8 And even if women are equally qualified, they may still perceive them- selves as being less qualified than men and therefore chose not to run (Lawless and Fox, 2005).

While these motivations are likely salient in many contexts, they also have important limitations. For example, we know that some portion of the observed electoral penalty that women receive is due to party nomination strategies, including women’s poorer placement within lists and dis-

7See, for example, Eagly et al. (1992); Beaman et al. (2009). 8Compounding this issue is the “double-bind” that women often face, as they need political experience but also have to juggle the reality that voters and parties prefer female politicians who are also “a wife and mother” (Teele et al., 2018).

5 tricts (Kelley and McAllister, 1984; Thomas and Bodet, 2013). Similarly, other work has found that female candidates are not significantly less qualified or prepared for office (Franceschet and Piscopo, 2012; Clayton, 2015) or that differences in experience only account for part of women’s unfavorable placement within lists or across districts (Esteve-Volart and Bagues, 2012).

Thus, although bias and candidate quality may play a role in shaping parties’ strategic behavior, we argue their response to quotas is also driven by a third and distinct factor: a preference for avoiding the displacement of male leaders in political networks.9 Because the allocation of a fixed num- ber of candidate places is typically zero-sum, each additional spot allocated to a female candidate is one less spot that can be allocated to a male candidate. As office-seeking politicians, (male) lead- ers are unlikely to want to give up their positions of power, such as desirable spots at the head of the list or in politically important or prestigious districts. Furthermore, in closed-list PR systems, the selection of candidates and determination of their rank also offers an opportunity for parties to use list spots to reward longtime supporters or to attract new talent and expand party mem- bership (Ferree, 2010). Ceding some of these spots to women may represent a loss in resources to build or maintain the support traditional networks and power brokers, who are predominantly men (Crowder-Meyer, 2013).

Recent literature has demonstrated the importance of male-leader- or incumbent-preserving be- havior in parties’ strategic response to quotas in a number of contexts. In Spain, for example, Esteve-Volart and Bagues (2012, p. 388) argue that women are “nominated according to how their presence in the list affects male candidates,” and that parties’ commitment to gender equality pre- vails only so long as it “does not cost a male candidate his seat.” As a result, parties manipulate senate lists to place women at the bottom when they expect to win only a handful of seats (pre- serving winnable seats for male candidates), and at the top only when they expect to win all seats. Indeed, the anticipated loss of seats for may be a driving factor in parties’ opposition to quotas in the first place. In Morocco, Sater (2012) notes that legislators feared losing seats for male in- cumbents under a proposed quota in 2002, and therefore ensured that the quota would expand the number of parliamentarians to include additional seats for women. Thus, while bias against women and supply-side issues are important, we should expect independent effects of the desire to maintain male leadership on parties’ strategic response to quotas.

3 Tunisia’s Gender Quotas and the 2018 Election Like many other countries, Tunisia has a long history of male-dominated politics and political networks (Benstead et al., 2015; Shalaby, 2016). Despite notable progress on women’s rights and access to health, education, and employment in the post-Independence period, the top-down ap- proach of “state-sponsored feminism” pursued by the Bourguiba and Ben Ali regimes did not

9Although sometimes conflated with party bias against women, this behavior is rooted not in gender per-se, but in the fact that parties seek to maintain existing leaders and networks, which are most likely to be dominated by men when quotas are first implemented.

6 fully incorporate women into public life (Charrad, 1997, 2001; Ennaji, 2016).10 However, ahead of the 2018 municipal elections, Tunisia adopted some of the most progressive and strict gender quotas in the world. Building on previous quotas for national-level elections in 2011 and 2014, the quotas first introduced for the municipal elections can be considered an “ideal type” as they are legislated, involve a closed-list PR system, require that 50 percent of candidates are women, have placement mandates, and are strictly enforced with a penalty of disqualification. Furthermore, unlike the law for Tunisia’s national-level elections, the municipal quotas impose additional re- strictions to ensure that an equal number of women become list heads.

The Tunisian 2018 municipal elections therefore present a unique opportunity to study the behav- ior of parties who are highly motivated to avoid displacing male candidates within an institutional environment that offers little room for maneuver. Although we expect parties’ strategies to adapt to quotas over time, it provides a baseline indicator of how parties might react to the initial shock under strict quota conditions.

3.1 Quota Adoption and Design Since the 2011 revolution, Tunisia has taken a number of important steps to deepen its democracy, including the introduction of electoral quotas designed to boost the participation of marginal- ized groups in positions of political leadership, including quotas for women, youth, and persons with disabilities. Under the country’s closed-list PR system, parties, coalitions, and independents present lists of candidates equal to the magnitude of the district in which they are running. For the 2011 National Constituent Assembly (NCA) elections and the 2014 elections for the parlia- ment (Assembly of the Representatives of the People, or ARP), 50 percent of these list spots were reserved for women. In addition, these elections included a placement mandate that required the lists to alternate between male and female candidates (Belschner, 2018). This “vertical parity” re- quirement was designed to help avoid some of the manipulations observed in other countries, where parties complied with the quota percent but simply placed female candidates at the bottom of the list. These quotas were strictly enforced.11

However, the national parliamentary quotas were not fool-proof. In particular, they did not im-

10Tunisia is often recognized for having the most progressive family law (the Code of Personal Status) in the region. At the same time, women’s rights have been highly politicized and part of a deliberate state-building strategy de- signed to win the allegiance of both internal and external actors (Dalmasso and Cavatorta, 2010; Tripp, 2019). Although some gender quotas did exist under the previous authoritarian regime as part of this strategy—e.g., the hegemonic Democratic Constitutional Rally (RCD) party decided to raise the number of women running on its list to 30 percent in 2007—these were not legislated or universal. Furthermore, Goulding (2009) argues that these quotas simply reproduced and entrenched existing power dynamic into a “patriarchal bargain.” 11In the national elections in 2011 and 2014, lists that did not comply with the gender quota were disqualified. Tunisia has also used youth quotas since 2011. For the 2018 municipal elections, a candidate under 36 had to be in one of the top three spots on the list and included every six candidates after that. Lists that did not comply were disqualified. A disability quota was first introduced in 2018 and requires one candidate with a disability to be included in the top ten spots of each list. The sanction for non-compliance is no reimbursement of campaign finances (Tavana and Russell, 2014; European Union, 2018).

7 pose any regulations on which gender was at the head of the list. In Tunisia’s fragmented party system, the sheer number of lists running reduced the anticipated number of seats for any indi- vidual list, meaning that often only the head of the list could be expected to win (Storm, 2014).12 As a result, 93 percent of lists in the 2011 election and 89 percent of lists in 2014 were headed by men (NDI, 2015). Thus, although women constituted 50 percent of candidates, most were not in a position to win the first—and often only—seat awarded to their list. Therefore, women only held around 30 percent of seats in the NCA after the 2011 elections and nearly 35 percent of seats in the ARP after the 2014 elections.

To address this shortcoming, the municipal elections law required vertical and horizontal par- ity. Under the new law, lists not only had to alternate between men and women, but 50 percent of each party’s lists must be headed by a woman (European Union, 2018). The adoption of this measure was the result of, not only sustained activism from Tunisian women’s and civil society organizations, but also support from most major parties and coalitions who believed they had an advantage in quota compliance over smaller parties. While some expressed concerns that im- plementing both the vertical and horizontal quotas in municipal elections would be difficult for smaller parties and independents given their smaller networks and capacity (Bin Hussein, 2016), the new law passed.13

3.2 Political Parities in the 2018 Elections In total, 2074 lists ran in the elections across 350 municipalities. Of these, 1214 were party or coali- tion lists and 860 were independent lists not formally linked to any political party. Of the parties which competed in 2018, there are two main categories: (1) the major parties, including the En- nahda Movement, a party with an Islamist orientation that ran lists in all 350 municipalities, and Nidaa Tounes, a big tent secular party that ran lists in 345 municipalities; and (2) smaller parties and coalitions that fielded a total of 519 lists, most competing in only a handful of municipalities. This second category—referred to as “third parties” throughout this article—included parties like Democratic Current and the People’s Movement and coalitions like the Popular Front.

As shown in Figure 1, Panel 1, both the main parties and smaller parties complied with the quota requirements of 50 percent FHLs. Ennahda and Nidaa had parity in the gender of their list heads, 48.2 percent of third party lists were headed by women (this 1.8 percent gap is due to the fact that parties that ran in an odd number of municipalities could have an additional list headed by a man). In comparison, independent lists that were not subject to the horizontal parity requirement14

12In 2014, for example, over 100 small and medium parties and independent lists ran in the parliamentary elections, in addition to the two main parties. 13The passage of the law with the support of major parties fits with our expectations that male politicians will extend rights to women when it is politically expedient (Teele, 2018) and that incumbent parties will adopt quotas when it gives them an advantage (Weeks, 2018). 14By definition, these lists only run in one municipality, and can therefore chose the gender of their list without the need to balance across municipalities.

8 overwhelmingly had male-headed lists (MHLs). In total, only 31 independent lists (3.6 percent) were headed by women. However, while it appears that parties largely adhered to the quota rules, we can see a clear drop-off in support for these FHLs at the ballot box. In total, only 72 (20.5 percent) of the lists that received the most votes in each municipality were headed by women, placing them at a disadvantage in the mayoral negotiation process.

50.7 Heads of lists 96.4 71.8Heads of winning lists100

Gender Lists 49.3 (%) Male Female 28.2

3.6 0 Party Independent Party Independent

List type

Figure 1: Gender of all list heads vs. first-place list heads by list type

3.3 Mayor Selection Unlike the direct election of municipal councils, mayors are indirectly elected by the winning councillors in the weeks directly following the election. Only the heads of the electoral lists are el- igible to run, and winning candidates must receive a simple majority of the support of the council members.15 Though lists that come in first in the municipality have an advantage going into the negotiations—particularly if that list has already secured a simple majority outright—it is not a forgone conclusion that the head of the first ranked list will become the mayor.

The negotiations for mayor occur behind closed doors; there is no public record of which council members supported which mayoral candidate. Based on our interviews, however, we know that the mayor position is highly sought after because of the power and prominence associated with the position. In all of our interviews, there were multiple list heads vying for the position. Addition- ally, list members tend to vote together as blocks for a mayoral candidate. Thus, the negotiations and jockeying for the position occur between the various parties and electoral lists. Sometimes these discussions occur between candidates directly, while other times they are mediated by the party elites and regional coordinators for the parties.

15If the mayor resigns, all council members can compete to replace them.

9 Ennahdha 42 89

Independent 119 Gender List type Male Nidaa 19 58 Female

Third Party 7 16

0 25 50 75 100 125

Number of mayors

Figure 2: Gender of list head and mayor by list type

As shown in Figure 2, women were elected mayor in only 68 out of 350 municipalities (or 19.4 per- cent of mayors). In part, this inequality is driven by the fact that few independent lists were headed by women. However, the proportion of women who become mayor on Ennahda and Nidaa lists is also less that the 50 percent anticipated by horizontal parity, indicating that other factors con- tributed to this phenomenon, including parties’ strategic placement of FHLs.

4 Research Design and Data Although Tunisian parties were relatively constrained in the strategies available to them given the closed-list PR system and strict placement mandates, we argue that they had significant lee- way to influence two elements of the electoral process that affected women’s representation: (1) the geographic placement of FHLs versus MHLs, and (2) the indirect election of mayors. A third strategy—running as independent lists—primarily applies to smaller parties and is examined in detail in another paper.16 We theorize that parties will seek to manipulate these processes pri- marily out of a concern with preserving the positions of male leaders, at the expense of female candidates.

4.1 Hypotheses The head of the list in Tunisian politics is an important symbol, a key part of the campaign, and the list member who is most likely to become mayor. In negotiating where to place FHLs versus

16Parties may chose to run as independent lists to circumvent the horizontal parity rule entirely. Because Ennahda and Nidaa lists ran in nearly every municipality, this phenomenon is likely isolated to smaller parties. However, anec- dotal evidence suggests that male candidates from female-headed Ennahda and Nidaa lists may also have defected and created their own parallel independent lists to run in the top spot.

10 MHLs, we argue that parties will seek to place FHLs in municipalities where the party has his- torically performed poorly. This expectation comes not only from the assumption that parties are motivated to reserve the most desirable positions for men (Esteve-Volart and Bagues, 2012), but also that party strongholds are more likely to have entrenched and powerful male party leaders and networks that may be more difficult to displace. In addition, we expect that parties will re- serve the most important (i.e., largest) municipalities for male leaders, placing FHLs in smaller and less prestigious locations (Lassebie,´ 2020). Together, this gives us two primary hypotheses:

Hypothesis 1 FHLs are more likely in municipalities where the party is relatively weak than MHLs.

Hypothesis 2 FHLs are more likely in smaller municipalities than MHLs.

Notably, these predictions run contrary to the observable implications if parties were primarily motivated by addressing voter bias or the lack of qualified female candidates. In these cases, we would expect FHLs in both “sure lose” and “sure win” municipalities, but MHLs in competitive ones (Frechette´ et al., 2008; Esteve-Volart and Bagues, 2012); similarly, we would expect FHLs in municipalities with the most progressive political cultures and greatest supply of qualified women, which are likely to be the largest municipalities concentrated in more urban areas.

Although the head of any list can run for mayor, lists with the most votes have an advantage. However, as a direct result of placing FHLs in areas where the party is relatively weak, we expect that FHLs will perform worse on election day. This leads to a third hypothesis:

Hypothesis 3 FHLs receive fewer votes and seats than MHLs.

Given that FHLs receive fewer votes and seats, it is likely that they will be in weaker positions going into the mayoral election process. While women whose lists came in second or third place could still be elected mayor, it is more likely that parties would instead take this opportunity to preserve the favored position of male leaders whose lists ranked higher. Furthermore, we expect that parties will support men for mayor even if their lists received fewer votes than an FHL. Therefore, we hypothesize that:

Hypothesis 4 Regardless of the share of votes received, women list heads are less likely to be elected mayor than men list heads.

Finally, we hypothesize that these party strategies are motivated by a desire to favor more politi- cally influential individuals, primarily men. Thus, we expect that:

Hypothesis 5 Male candidates are more politically connected than female candidates.

11 4.2 Data We employ a multi-method approach to test our theory, drawing on both list- and candidate- level data. We also leverage an original survey and interviews with candidates, elected council members, and party officials to provide additional evidence on the potential motivations for party behavior.

Election and Municipal Data. To examine the strategic placement of FHLs, we use a combination of data from the 2018 municipal elections, the 2014 parliamentary elections, and municipal charac- teristics. Our primary dependent variable—whether or not a list has a female head—comes from data on the lists and candidates competing in the 2018 elections published by the Tunisian electoral commission (ISIE) on their website.17 To code the gender of each of the 2074 list heads, we used a dictionary of Tunisian Arabic names developed for this project.18

The primary independent variables used to predict FHL placement include measures of each party’s relative strength in a given municipality, as well as the log municipal population from the 2014 Census. Since we need a measure of relative strength that predates the 2018 election, we construct the strength measure by aggregating each party’s vote share in the 2014 parliamentary elections to the municipal level, and then subtracting the mean of that party’s vote share across all municipalities.19

While other studies analyze the competitiveness of the race—typically measured as raw vote share or the margin of difference between parties (Thomas and Bodet, 2013; Ichino and Nathan, 2013; Harbers, 2014)—focusing on the performance of the party in a given municipality relative to its performance in other municipalities is a more appropriate measure for our theory since parties’ placement decisions depend on trading off the costs or benefits of an FHL across municipalities where they are running. This measure of relative strength also allows us to better control for trends in national performance across parties (e.g., Third Parties have much smaller vote shares than En- nahda and Nidaa). However, because our measure of strength is based on 2014 data, our analysis includes only those parties who ran in 2014 and 2018.20 To allow for better interpretation of model coefficients, we standardize this variable and other continuous measures in all of our analyses.

We also use the 2014 election data, the 2014 census, and data from the Ministry of Local Affairs to construct a number of controls, including the percent of women living in the municipality who

17See http://isie.tn. 18To create this name-gender dictionary, we used our knowledge of common Tunisian names and the fact that lists alternate candidates by gender to code unknown names. This dictionary was validated by a native speaker. 19To construct this and other measures from the 2014 election results, we digitized the 2014 data using PDFs of hand- written results available on the ISIE website. Due to missing returns on the ISIE website, there are ten municipalities with missing data on party strongholds and female turnout in 2014. See Supplementary Appendix for details. 20It is possible that some parties from 2014 chose not to run in 2018 or chose to run as independents instead. If this is the case, it means that the parties in our analysis self-selected into running and complying with the quotas, which suggests a more difficult test of our theory.

12 voted in 2014 (female turnout) and the municipality type (whether old, new in 2015, or expanded in 2015).21 We use additional data to test for alternative hypotheses, including the percent of women and men with a college degree, and the percent of women members of special delegations, which were the appointed municipal councils put in place after the revolution that governed until the 2018 elections.22

Finally, to determine whether there is less support for women to become mayors than men, we constructed a data set with the gender and party affiliation of each mayor elected in 2018 in Tunisia’s 350 municipalities. All of the negotiations over the indirect elections for head of the municipal council are private, but newspaper reports covered the inauguration of all 350 mayors.

Survey Data. We use data from two surveys conducted in Tunisia in 2018. First, we use data from an original survey of approximately 1900 candidates for the elections. We conducted this Local Election Candidate Survey (LECS) between April and May 2018 across 100 municipalities and included questions that measure a variety of candidate characteristics, attitudes, and polit- ical experiences.23 We merge candidate responses with data on the electoral and municipality characteristics to test alternative hypotheses on candidate quality and attitudes towards women’s leadership.

Given that the public profile of many of the candidates for municipal elections is low, the LECS enables us to collect measures of candidate quality, previous political experience, and embedded- ness in political networks. These measures include candidate education, political ambition, polit- ical knowledge, skills (e.g., public speaking experience), political activities, membership in local organizations such as unions, and previous political experience. In addition, the LECS includes a conjoint experiment that measures the relative importance of gender in candidates’ evaluations of potential running mates.24

Second, we leverage a nationally-representative survey of over 6500 Tunisian citizens fielded di- rectly following the 2018 municipal elections. The nationally-representative survey data were col- lected by Democracy International in the month directly following the municipal election. It asked citizens about their voting behavior during the municipal elections, as well as attitudinal mea-

21Before 2015, Tunisia had 264 municipalities, which did not cover rural areas. In 2015, the Ministry of Interior announced that municipalities would be created or expanded to cover the entire territory. This reorganization created three categories of municipality: (1) old municipalities, (2) new municipalities that consist of primarily rural sectors, and (3) so-called “expansion” municipalities that consist of an old municipality expanded to include rural sectors. 22Education data comes from the 2014 census but is only available at the delegation level, which we match to munic- ipalities. Data on the gender of special delegation members comes from compiling information on the names of special delegation members between 2011-2018 from the Tunisian Official Gazette (JORT) and running these names against the gender name dictionary. 23Further details about the sampling strategy and design of this survey are available in the Supplementary Ap- pendix. The survey is representative of municipalities with different sized councils, and top-ranked candidates from the two main parties—the Ennahda Movement and Nidaa Tounes. It also includes a substantial number of candidates from independent and other party lists and some lower-ranked candidates. 24These questions are provided in detail in the Supplementary Appendix.

13 sures toward gender egalitarianism and secularism that we use to construct indices measuring local popular attitudes.

Interviews. Prior to the elections, we conducted several dozen interviews with potential candi- dates and local party and civil society activists to inform our hypotheses and survey design. Fol- lowing the election, we selected a set of municipalities in which to conduct interviews with the mayors themselves. We carefully chose these municipalities to ensure a diverse interviewee pool. We included mayors elected on the Nidaa Tounes, Ennahda, and independent lists from three governorates: Tunis, Monastir, and Tataouine, as well as mayors elected on the Democratic Cur- rent and People’s Movement lists in Manouba, Beja, and Bizerte.25 For each party, both male and female interviewees were selected. These sixteen semi-structured, post-election interviews with mayors give us additional evidence regarding the logic of parties’ strategic placement of FHLs and the mayoral selection process.

4.3 Estimation Strategy To examine whether Tunisian parties place FHLs in relatively weak and small municipalities com- pared with MHLs (H1 & H2), we use the following equation to estimate the likelihood that a party list was headed by a woman:26

FHL2018lmg = β1Strength2014lmg + β2Popmg + β3Partyl + β4Xmg + αg + ulmg (1) where FHL2018lmg is a dummy variable equal to 1 if party list l in municipality m had a female head; Strength2014lmg is the party’s relative strength in the municipality in 2014; P opmg is the log of municipal population in 2014; Partyl is the list’s party; Xmg is a vector of municipal characteristics, including 2014 female turnout and municipal type; and αg is governorate fixed effects. Standard errors (ulmg) are clustered by municipality m to account for correlated errors across lists running against each other.

Using LECS data for candidates in a sample of 100 municipalities, we then estimate an individual- level version of the above model to predict whether a candidate becomes the list head that also controls for candidate-level quality and experience measures:

ListHeadilmg = β1Femalei × β1Strength2014lmg + β2Popmg + β3Partyl + β4Xi + αg + uilmg (2) where ListHeadilmg is a dummy variable equal to 1 if candidate i is the head of list l in munici- pality m; Femalei is a dummy variable equal to 1 if the candidate is female; Strength2014lmg is the party’s relative strength in the municipality in 2014; P opmg is the log of municipal population in

25Candidates from the Democratic Current and People’s Movement lists were not chosen as mayors in any munici- palities Tunis, Monastir, and Tataouine Governorates. 26For ease of interpretation, the tables included in this article show ordinary least squares (OLS) models, however, logit models are used for graphing and tables are also available in the Supplementary Appendix.

14 2014; Partyl is the list’s party; Xi is a vector of individual characteristics, including age, education, political experience, skills, ambition, and knowledge; and αg is governorate fixed effects. Standard errors (uilmg) are clustered by municipality m. To increase the validity of this model, this analysis is limited to first and second-place candidates27 from lists included in LECS, as these are most likely to be considered for the list head position. Alternate specifications also interact candidate gender with population, and split the sample into male and female candidates.

Next, we explore the consequences of these strategies for list performance in the 2018 municipal elections and the subsequent selection of mayors from among the elected councillors. To begin, we predict the vote share, list rank, and seats won for each list (H3):

Outcomelmg = β1FHLlmg + β2Partyl + β3CouncilSeatsmg + β4NumListsmg + αg + ulmg (3) where Outcomelmg is alternately the percent of votes, rank based on vote share, and the total seats won by list l in the 2018 elections in municipality m; FHLlmg indicates a female-headed list; P artyl is party; CouncilSeatsmg is the number of seats in the municipal council; NumListsmg is the number of lists running in the municipality; αg are governorate fixed effects; and clustered standard errors.

To test whether women were elected mayor at the same rate as men (H4), we estimate the follow- ing equation:

WonMayorlmg = β1FHLlmg + β2Performancelmg + β3NumListsmg + β4ListTypel + αg + ulmg (4) where WonMayorlmg is a dummy variable equal to 1 if list l’s head won the mayorship in munici- pality m; FHLlmg indicates a female-headed list; Performancelmg is either the percent of municipal votes received in 2018 or the list rank based on vote share; NumListsmg is the number of lists run- ning in the municipality; ListTypel is the list type (Ennahda, Nidaa, third party, or independent); (αg) is governorate fixed effects; and ulmg are municipality-clustered standard errors.

Finally, to explore the motivations behind party strategy and test alternative hypotheses, we use variations on models described above with additional controls, additional data from the LECS— including a conjoint experiment and a series of questions measuring candidates’ qualifications and embeddedness in existing political networks—and interview data.

5 Main Results: Predictors and Consequences of Female-Headed Lists 5.1 Strategic placement of FHLs Table 1 provides the results of Equation 1. We find clear evidence that Tunisian political parties did not place FHLs at random. Party strength—i.e., parties’ performance in the municipality during

27Due to the alternating gender list construction, one of the top-two list candidates will be male and the other female.

15 the 2014 elections relative to their national averages—is significantly and negatively correlated with the probability of having an FHL. Consistent with H1, we find that a one standard devia- tion increase in a party’s relative strength decreases the probability of an FHL by approximately 4.3 percentage points (Model 1). Overall, a party running in a municipality where its 2014 perfor- mance was above its national average was 7.3 percentage points less likely to run a FHL (Model 3). The non-significant squared term added in Model 2 suggests that the relationship between party strength and FHL placement is relatively linear, rather than parabolic as we might expect if parties were placing women in both sure-win and sure-lose municipalities.

Table 1: Strategic Placement of Female-Headed-Lists (FHLs)

Dependent variable: Probability of an FHL Linear Quadratic Binary (1) (2) (3) Relative strength in 2014 −0.043∗∗∗ −0.056∗∗∗ (0.015) (0.018)

Relative strength in 2014 squared 0.031 (0.023)

Above mean relative strength in 2014 −0.073∗∗ (0.030)

Log population −0.120∗∗∗ −0.119∗∗∗ −0.120∗∗∗ (0.018) (0.018) (0.018)

Female turnout in 2014 0.055∗∗ 0.052∗∗ 0.054∗∗ (0.025) (0.025) (0.025)

Constant 0.474∗∗∗ 0.485∗∗∗ 0.510∗∗∗ (0.098) (0.098) (0.102)

Controls Y Y Y Governorate FE Y Y Y F Statistic 4.427*** 4.42*** 4.289*** Observations 988 988 988 R2 0.110 0.112 0.108 Adjusted R2 0.082 0.083 0.080 Residual Std. Error 0.479 (df = 957) 0.479 (df = 956) 0.480 (df = 957) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 OLS models with standardized continuous variables (2014 relative strength, log population, and female turnout) and standard errors clustered by municipality that also control for party and municipal type. Relative strength is the party’s voteshare in the 2014 parliamentary elections aggregated to the municipal level, minus the party’s average voteshare across all municipalities. Analysis excludes independents lists and parties or coalitions that ran in 2018 but not 2014.

As shown in Table 1 and Figure 3—and consistent with H2—we also find that FHLs are less likely in larger municipalities. For every one standard deviation increase in log population size of a municipality, the probability of an FHL decreases by approximately 12 percentage points. For ex- ample, in a municipality like Lala (Gafsa Governorate) with a population of around 10,000 people (-1 s.d.), the predicted probability of an FHL is approximately 67 percent, compared with 41 per- cent in a municipality like Menzel Bourguiba (Bizerte Governorate) with 60,000 people (+1 s.d.). Finally, we see that parties were more likely to place FHLs in municipalities that had a higher

16 level of female turnout in the 2014 parliamentary elections. These results are robust to a variety of alternative measures, specifications, and sub-samples (see Supplementary Appendix).

100%

75%

Log population Predicted probability 50% High (+1 SDs) of FHL Medium (0 SDs) Low (−1 SDs)

25%

0% −2 −1 0 1 2 3

Relative strength in 2014 (standard deviations)

Predicted values based on a logit model controlling for party, women's turnout in 2014, municipal type, and with governorate fixed effects.

Figure 3: FHL Placement by Relative Strength

In addition, these patterns generally hold when we look within lists using the LECS data.28 Figure 4 shows the predicted probability of becoming the list head after controlling for characteristics such as their age and education, political experience, knowledge, ambition, and relevant skills (Equation 2). Here, we see that in areas where the party had below-mean performance in the mu- nicipality in 2014, the predicted probability that the female candidate was selected as list head is approximately 70 percent, compared with a 39 percent probability for the male candidate. In ad- dition, top female candidates are more likely to be the list head in small municipalities, compared with large ones.

28The interaction between the relative strength indicator variable and candidate gender loses significance when standard errors are clustered at the municipal level; however, it remains a significant predictor for women candidates when the sample is split. The interaction between gender and population size is significant at the 95 percent level. For full results, see the Supplementary Appendix).

17 100% 100%

75% 75%

Probability Candidate Candidate of being 50% Female 50% Female list head Male Male

25% 25%

0% 0% Below mean Above mean −2 0 2 4

Party's relative strength Log population (standard deviations) Predicted values based on a logit model controlling for candidate characteristics, including age, education, political experience, skills, knowledge, and ambition, as well as list type, log of municipal population, and governorate fixed effects. Sample is restricted to the top two male and female candidates on lists interviewed for the LECS. Graph shows 95 percent confidence intervals.

Figure 4: Individual-Level Predictors of Becoming List Head

Overall, these results indicate that the parties are placing female- and male-headed lists strategi- cally, with a party’s FHLs more likely in municipalities in which the party has historically per- formed poorly and in those with less power and prestige.

5.2 Electing Mayors We can see the result of these placement strategies in Figure 5, which estimates the electoral penalty for FHLs in the 2018 elections, in terms of the votes they received and the resulting rank and number of seats won by their list (Equation 3). Consistent with H3, FHLs received approxi- mately 6.8 percentage points fewer votes than those headed by men, which translated into a worse overall rank for FHLs (second place rather than first). Furthermore, because FHLs won an average of one fewer seat, winning female list heads had fewer co-partisans in the council to support their candidacy for mayor.

18 Vote share Rank Seats Won 8

2.5 31.1 2.4 7.3 30 7 List head 2.0 Female Male 1.5 25 24.7 1.4 6 6

1.0

Predicted values based on OLS model controlling for party, number of lists running, number of municipal council seats, and governorate fixed effects. Shown with 95 percent confidence intervals.

Figure 5: Electoral outcomes for FHLs

Finally, as Table 2 demonstrates, similarly ranked female list heads secure the mayorship at lower rates than men, consistent with H4. Overall in Model 1, we see that women list heads are approxi- mately 3.3 percentage points less likely to become mayor than their male counterparts, controlling for list, the percent of votes received, list type, and the number of lists running in the municipality. However, this gap grows larger when we restrict the sample to lists with a reasonable expectation of winning the mayorship in Model 3.29 For the top-three ranked lists, FHLs were approximately 9.9 percentage points less likely to be selected mayor.

29Out of 350 municipalities, 326 chose mayors from the top three lists.

19 Table 2: Probability of Female List Head Becoming Mayor

Dependent variable: Probability of being elected mayor All lists Party lists Top 3 (1) (2) (3) FHL −0.033∗∗ −0.029∗ −0.099∗∗∗ (0.016) (0.018) (0.029)

Vote share (%) 0.021∗∗∗ 0.021∗∗∗ (0.001) (0.001)

Ranked second −0.424∗∗∗ (0.044)

Ranked third −0.518∗∗∗ (0.039)

Constant −0.345∗∗∗ −0.372∗∗∗ 0.780∗∗∗ (0.045) (0.064) (0.037)

Controls Y Y Y Governorate FE Y Y Y F Statistic 88.544*** 48.623*** 13.688*** Observations 2,074 1,214 1,044 R2 0.379 0.407 0.279 Adjusted R2 0.370 0.393 0.258 Residual Std. Error 0.297 (df = 2044) 0.306 (df = 1185) 0.399 (df = 1013) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 OLS models that also control for list type and the number of lists running, with standard errors clustered by municipality. Model 1 includes all lists, Model 2 includes all party and coalition lists, Model 3 includes only the top-three ranked lists. Results consistent with logit models, but OLS shown here for ease of interpretation.

6 Exploring Motivations for Party Behavior These results indicate that political parties in Tunisia strategically placed female-headed lists in weaker municipalities, leaving female candidates at a disadvantage in the mayoral selection pro- cess. In this section, we examine the possible explanations for this pattern. We argue that this strat- egy is consistent with the theory that political parties in Tunisia were primarily motivated by a de- sire to avoid displacing politically-connected male leaders and preserving important positions— including mayorships and control of councils in large municipalities—for men.30 While bias and candidate quality may have contributed to the strategy or played a role in select cases, they are insufficient to explain the observed results. We examine each of these potential motivations in turn.

30The most notable exception to this overall trend is Souad Abderrahim, who ran as the head of Ennahda’s list in Tunis—Tunisia’s capital and largest city—and was then elected as the city’s first female mayor. She has been one of the party’s most prominent female members since being elected as a representative to the National Constituent Assembly in 2011.

20 6.1 Male Dominance of the Political Sphere When facing new gender quota requirements, we argue that parties are concerned about pro- tecting the prerogatives of important and influential politicians and will engage in strategic list construction to avoid displacing these (traditionally male) leaders. While these individuals might be incumbents, they are not exclusively previous office holders and can include party activists or influential community members active in local political and associational life. Given the relative marginalization of women in positions of political leadership historically in Tunisia, we expect that men will be more embedded in existing political networks than women.31

To examine gendered differences in political embeddedness, we draw on six key measures in our LECS data: (1) party membership; (2) years in party; (3) activities index, including political activ- ities such as attending a municipal council meeting, attending a protest or sit-in, or contacting an MP in the previous year; (4) membership index, based on the candidate’s membership in a union, professional organization, civil society organization, business association, recreational or sports club, or religious organization; (5) previous member of the hegemonic Democratic Constitutional Rally (RCD) party under Ben Ali; and (6) previous political experience.32 We contrast the top male vs. female candidates—i.e., the first- and second-ranked candidates on party lists—along these six dimensions.33 Table 3 displays the difference in means between top male and female candidates.

Table 3: Difference in Means: Top Male and Female Candidates

Mean (Men) Mean (Women) Difference p.value Political Embeddedness Measures Party Member 0.729 0.733 -0.004 0.940 Years in Party 8.519 2.948 5.571∗∗∗ 0.000 Activities Index 2.359 1.665 0.694∗∗∗ 0.000 Membership Index 1.517 1.086 0.431∗∗∗ 0.001 RCD Member 0.194 0.061 0.133∗∗∗ 0.000 Political Experience 0.503 0.170 0.333∗∗∗ 0.000 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 Sample is limited to candidates ranked first or second on a party list (n=348).

As Table 3 indicates, the top men on parties’ lists score higher on average than the top women on five of the six measures of political embeddedness. Among candidates who are formally members of a political party, men have been members for over eight and a half years, over five and a half years longer than female party members on average. Men are also more likely to have engaged in political activities in the previous year and have higher rates of organizational membership, including membership in the hegemonic RCD party of the Ben Ali regime. Finally, men are more likely to have previous political experience, either in government office, on the municipal council, or in a previous election. Over 50 percent of top male candidates have previous political experience compared to only 17 percent of top female candidates.

31For a discussion of this pattern, see Benstead et al. (2015). These patterns are certainly not unique to Tunisia (Esteve-Volart and Bagues, 2012; Crowder-Meyer, 2013). 32The full details of these measures are in the Supplementary Appendix. 33The results hold when we look at the full sample.

21 This type of embeddedness in the existing political networks was critical for local party recruit- ment. In most cases, local and regional party officials led the process of list construction, and these offices often sought out local leaders, primarily male, to lead their lists. In an interview, a regional coordinator for Ennahda stated that, in the Sidi Bouzid Governorate, the party gave the choice of list head to the eight municipalities (out of seventeen total) in which the party had performed best in the governorate. All of them opted to have a male list head.34 The list heads were typically identified as a result of their political connections. For instance, the head of Nidaa Tounes’ list in one of the municipalities in the Monastir Governorate had previously served as the mayor of the municipality under Ben Ali, as well as in the national parliament. He reported that Nidaa’s regional coordinator in Monastir reached out to him about heading the party’s list because of his political history.35

6.2 Alternative Explanations Broadly, our main results are inconsistent with the theory that parties are motivated primarily by voter bias or the lack of qualified female candidates. In either case, for example, we would expect FHLs to be more likely in larger cities, as these are typically more ideologically progressive than rural municipalities and believed to have the highest supply of qualified female candidates (Abdo-Katsipis, 2017). To examine these potential motivations systematically, the remainder of this section explores these two alternative explanations for parties’ placement of FHLs.

Bias against female candidates. Recent research in Tunisia suggests that citizens’ gender attitudes can play a role in their support for male and female politicians (Benstead et al., 2015; Blackman and Jackson, 2019).36 We explore the potential influence of voter bias through interviews, sur- vey data, and a conjoint experiment. In interviews, several candidates and regional coordinators expressed concerns about voter bias. During the list construction period, local Ennahda coordina- tors in Moknine, a large municipality in Monastir, reported that they had requested Ennahda run a MHL in their municipality because “Moknine is a patriarchal society.”37 During and even after her campaign to become mayor, the mayor of a small municipality in the Monastir Governorate encountered men saying that a woman was not capable of running the municipal council.38

However, other interview evidence supports the view that local ideologies were not the main bar- rier to female political leadership. Several interviewees stressed the cooperation between men and women on electoral lists, with all playing a role in the campaigns. An independent mayor from the coastal governorate of Monastir emphasized that there were no differences between the work

34Interview, May 3, 2019. 35Interview, October 19, 2019. 36Conversely, Bush and Prather (2020) find no evidence that Tunisians perceive their elected female MPs to be more or less qualified or effective than their male counterparts. 37Interview, March 23, 2018. 38Interview, October 19, 2019.

22 of men and women during the campaign.39 A female mayor in the Tataouine Governorate stated that she was able to go door-to-door and attend meetings like her male counterparts.40 Souad Ab- derrahim, the first female mayor of Tunis, reported that she has been embraced by voters: “Men and women, people from different parties are happy and proud to have a woman leading the capital city.”41

Moreover, in our interviews, concerns about voters’ perceptions of female political leaders did not follow a clear pattern in terms of region, level of urbanization, or municipality traits. Candidates in large, urban municipalities located in Tunisia’s coastal regions expressed some of the same con- cerns about voter bias as candidates in rural municipalities or those in the South or Interior that are typically considered to be more conservative. Overall, this suggests that, while some candi- dates were concerned about voter bias, not all candidates shared that concern, and it did not have a clear effect on the placement of FHLs; while party elites may have recognized or articulated this concern, it does not appear to be the driving force behind parties’ strategic behavior in response to the gender quotas.

Building on our interviews, we test the potential role of voter bias by examining the robustness of our main results to the inclusion of measures for the conservatism of local voters, both in terms of popular attitudes regarding secularism and gender egalitarianism. Table A.7 in the Appendix provides additional evidence that placement of FHLs is not primarily driven by the ideology of local voters. The coefficient on the relative strength of a party within a municipality is stable and significant, even after controlling for additional measures of ideology in the local population that might increase or decrease voter bias against FHLs, including secular attitudes (Model 2) and feminist attitudes (Model 3).42 The coefficients on public attitudes toward secularism and gender egalitarianism also do not indicate that local-level popular attitudes played a clear role in FHL placement decisions.

In addition to voter bias, there may be concerns that party elites are biased against potential female candidates. To test this potential motivation, we examine whether the results are driven by parties with a particular ideology43 and whether candidates themselves are resistant to the inclusion of women on electoral lists. Model 1 of Table A.7 examines the results by party. While we do see some evidence that strategic placement varied by party and was stronger for Ennahda than Nidaa lists, the fact that Third Parties—the largest of which were left-leaning parties and coalitions such as

39Interview, October 19, 2019. 40Interview, October 23, 2019. 41Interview, October 25, 2019. 42Using the survey of over 6500 Tunisians fielded by Democracy International in June 2018, we construct municipality-level measures of secularism and pro-women attitudes. See Supplementary Appendix for details. 43For instance, some critics of Islamism might argue gatekeepers of the Islamist Ennahda Movement are more biased against women than other parties and therefore sought to keep them out of power for ideological reasons. We find no evidence to support this. At both the mass and elite levels, attitudes toward gender egalitarianism are not correlated with partisanship.

23 Popular Front and People’s Movement—also placed FHLs where they had historically performed the weakest indicates that this is not a story about ideology.44

Moreover, the LECS data provide additional evidence that candidates themselves are not inher- ently biased against female politicians more broadly. To test general attitudes toward female candi- dates, we included a conjoint experiment in our candidate survey that asked respondents to select their preferred list mate from among two options, with hypothetical candidate attributes varying based on gender, age, profession, length of party membership, and background.45 Given the ran- domization of the attributes, we can identify the average marginal component effect (AMCE) for each attribute’s levels.

Figure 6: Candidate preferences for potential list mates

Figure 6 displays the AMCE results and 95 percent confidence intervals for the hypothetical can- didate’s gender for all respondents (Panel 1) and by respondents’ party affiliation (Panel 2). Full results for all attributes are displayed in Figures A.5 and A.6 in the Supplementary Appendix. We find that candidates running in the municipal elections do not display specific bias against potential female running mates. These findings indicate that the strategic placement of female

44We explore variation in specific parties’ placement strategies in detail in another paper. 45Further details on the conjoint experiment are included in the Supplementary Appendix. We did not ask candi- dates who they would prefer as the head of the list. For additional details on using conjoint experiments to assess preferences regarding political candidates, see Hainmueller et al. (2014) and Schwarz and Coppock (2020).

24 candidates is not driven by a general bias against female candidates or a bias among members of any one party. However, this does not mean that parties are supportive of female leadership, as illustrated by the strategic placement of FHL and the reduced support for winning female list heads in the mayoral elections.

Supply of qualified female candidates. Another potential explanation for the strategic place- ment of FHLs is the under-supply of qualified female candidates. We explore the potential role of supply-side issues by examining the robustness of our main results to the inclusion of proxies for the local supply of female candidates and by examining differences between potential list heads on a variety of quality measures.

Table A.7 in the Appendix includes two tests of the supply-side theory. First, we find no signifi- cant relationship between the percent of female special delegation members—the appointed local councils that governed municipalities between 2011 and 2018—and whether a list has a female head (Model 4).46 Second, Models 5 and 6 show that, although a one standard-deviation increase in the percent of women in the municipality with a secondary education or higher is associated with a 7 percentage point increase in the likelihood of an FHL, there is a similar effect associated with a higher percentage of men with a secondary education or higher. These results suggest that FHLs are more likely in municipalities with more educated adults overall. The coefficient on the relative strength measures is robust to the inclusion of these variables.

In addition, the LECS data allow us to look more deeply at the relative quality of the top male vs. female candidates (Figure 7), as well as the top female candidates across male- and female-headed lists (Table 4). We examine five measures of candidate quality, including: (1) education level; (2) political ambition; (3) political knowledge; (4) skills index, based on whether the candidate had ever engaged in public speaking, policy research, fundraising, recruitment, or event planning; and (5) previous political experience.47 Measures like previous political experience reflect, at least to some degree, the gatekeeping by powerful men discussed above rather than quality, but, because it is often used as a measure of candidate quality, we include it here for completeness.

46Special delegation members are well-positioned to run in the municipal elections because they have previously worked on the council and may be familiar to voters. 47The full details of how we constructed these measures are in the Supplementary Appendix.

25 2.0

1.5

Gender Outcome 1.0 mean Female Male

0.5

0.0 Education Ambition Knowledge Skills Experience

Candidate quality measure

Graph shows averages of candidate quality, comparing the first− and second−ranked candidates on interviewed party lists (n = 348). Data comes from the Authors' Local Election Candidate Survey (LECS) of candidates for the 2018 municipal elections. 'Education' is a dummy for attended university; 'Ambition' is a dummy for desire to run for parliament in the future; 'Knowledge' is a dummy for identifying the correct number of municipal council seats; 'Skills' is an index of relevant policy skills; and 'Experience' is a dummy of whether the candidate had previously run in an election or served on the local council or in government administration. For more details on the LECS survey, see Appendix.

Figure 7: Contrasting Top Male and Female Candidates

Among candidates on party lists, top-ranked female candidates are on average more educated and express greater political ambition than top-ranked males. Top-ranked men, by contrast, are more likely to have previous political experience and are more likely correctly identify the number of seats on their local municipal council. There are no significant differences between men and women on the Skills Index. Furthermore, and as shown in Figure 4 previously, women are still less likely to be chosen as list head in party strongholds or important municipalities even after controlling for political experience and other quality measures. Importantly, men do not clearly dominate the measures of quality. Taken together, there is little evidence that the placement of FHLs is driven by the absence of qualified female candidates.

Additionally, when we look at the top female candidates across lists headed by men and women (i.e., the first-ranked candidate on an FHL, and the second-ranked candidate on an MHL), we find no statistically significant differences in candidate quality (Table 4). This provides further evidence that the supply of female candidates does not vary systematically between the municipalities in which parties placed FHLs and MHLs.

26 Table 4: Difference in Means: Top Women on FHL and MHL

Mean (FHL) Mean (MHL) Difference p.value Candidate Quality Measures University Education 0.924 0.943 -0.018 0.637 Political Ambition 0.649 0.667 -0.017 0.817 Knowledge 0.895 0.812 0.083 0.137 Skill Index 1.162 1.360 -0.198 0.274 Political Experience 0.218 0.126 0.092 0.124 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 Sample is limited to women ranked first or second on a party list. N=166

7 Discussion Tunisia’s 2018 municipal elections provide a rich case for deepening our understanding of the obstacles facing female politicians and of the strategic response of parties to new formal institu- tions, such as gender quotas. Following the 2011 revolution and national-level quota adoption, the Tunisian government strengthened these gender quotas for its first democratic municipal elec- tions, mandating horizontal, as well as vertical parity between male and female candidates. On paper, these quotas provide little opportunity for circumvention: they are mandated by law, well- enforced and require both that each electoral list alternate between male and female candidates and that half a given party’s lists are headed by women.

While parties competing in the 2018 municipal elections may have had limited ability to evade the quotas entirely, they had strong incentives to minimize the quota’s impact on male politicians embedded in their political networks. Using data from the 2014 national and 2018 local elections in Tunisia and an original survey of candidates, we demonstrate a set of behaviors consistent with this theory. We show that parties strategically placed female-headed lists (FHLs) in their historically weakest municipalities and in the smallest, least political important municipalities.

This strategy had several important implications. First, as a direct result of this strategy, women were less likely to obtain higher levels of political office. By concentrating FHLs in weaker, less im- portant districts, female candidates were at a disadvantage during the mayoral selection process and less likely to be selected as the mayor. Thus, while women make up nearly 50 percent of coun- cil members, they hold less than 20 percent of mayorships. In comparison to other countries, this is an impressive achievement. Still, understanding the dynamics that led to this gap in descriptive representation at high levels of political office is essential for ensuring the full inclusion of women in positions of power. Future work should also examine how these gender gaps at different levels of political office shape substantive representation and the responsiveness of different levels of government.

Second, this party strategy reinforces male dominance in the political sphere, particularly at higher levels. Increasing representation at the local level is important for the recruitment of candidates for higher-level office. In contrast to national legislatures that are typically on the order of hun- dreds of representatives, local elections have the potential to inject thousands or tens of thousands

27 of women into the political sphere, helping to create a more solid pipeline of experienced female politicians. Policy-makers should be attentive to these gendered patterns of recruitment and place- ment if diversifying who holds political power is the goal.

Finally, these results also have important implications for our broader understanding of parties’ strategic response to changing formal institutions, namely how entrenched political interests in- teract with new gender quota requirements. Even where quotas achieve one stated goal—equal representation among candidates or politicians—achieving gender parity in access to the most powerful positions likely requires additional measures, struggle, and time. Tunisia’s 2018 elec- tions demonstrate that, while quotas with certain design features may be successful in getting more women elected, they are only the first step in institutionalizing female leadership.

28 References Abdo-Katsipis, C. B. (2017). Women, political participation, and the arab spring: Political awareness and participation in democratizing tunisia. Journal of Women, Politics & Policy, 38(4):413–429. Barnes, T. and Burchard, S. (2012). Engendering politics: The impact of descriptive political representation on women’s political engagement in sub-saharan africa. Comparative Political Studies, 46(7):767–790. Beaman, L., Chattopadhyay, R., Duflo, E., Pande, R., and Topalova, P. (2009). Powerful women: does exposure reduce bias? The Quarterly Journal of Economics, 124(4):1497–1540. Belschner, J. (2018). The adoption of youth quotas after the arab uprisings. Politics, Groups, and Identities, pages 1–19. Belschner, J. and Garcia de Paredes, M. (2020). Hierarchies of representation: The re-distributive effects of gender and youth quotas. Representation, pages 1–20. Benstead, L. J. (2019). Do female local councilors improve women’s representation? The Journal of the Middle East and Africa, 10(2):95–119. Benstead, L. J., Jamal, A. A., and Lust, E. (2015). Is It Gender, Religiosity or Both? A Role Congruity Theory of Candidate Electability in Transitional Tunisia. Perspectives on Politics, 13(1):74–94. Besley, T., Folke, O., Persson, T., and Rickne, J. (2017). Gender quotas and the crisis of the mediocre man: Theory and evidence from sweden. American economic review, 107(8):2204–42. Bin Hussein, I. (2016). Controversy after the Tunisian parliament approved a law on “horizontal and vertical parity” (Arabic). Al-Araby, June 21, 2016. Blackman, A. and Jackson, M. (2019). Gender Stereotypes, Political Leadership, and Voting Behavior in Tunisia. Political Behavior. Braga, M. and Scervini, F. (2017). The performance of politicians: The effect of gender quotas. European Journal of Political Economy, 46:1–14. Bush, S. S. and Prather, L. (2020). Islam, gender segregation, and political engagement: evidence from an experiment in Tunisia. Political Science Research and Methods, pages 1–17. Charrad, M. (1997). Policy Shifts: State, Islam, and Gender in Tunisia, 1930s-1990s. Social Politics: International Studies in Gender, State & Society, 4(2):284–319. Charrad, M. (2001). States and Women’s Rights: The Making of Postcolonial Tunisia, Algeria, and Morocco. University of California Press. Chattopadhyay, R. and Duflo, E. (2004). Women as policy makers: Evidence from a randomized policy experiment in india. Econometrica, 72(5):1409–1443. Clayton, A. (2015). Women’s political engagement under quota-mandated female representation: Evidence from a randomized policy experiment. Comparative Political Studies, 48(3):333–369. Crowder-Meyer, M. (2013). Gendered recruitment without trying: How local party recruiters affect women’s representation. Politics & Gender, 9(4):390–413.

29 Dalmasso, E. and Cavatorta, F. (2010). Reforming the family code in tunisia and morocco–the struggle between religion, globalisation and democracy. Totalitarian Movements and Political Religions, 11(2):213–228.

De Paola, M., Scoppa, V., and De Benedetto, M. A. (2014). The impact of gender quotas on electoral participation: Evidence from italian municipalities. European Journal of Political Economy, 35:141–157.

Eagly, A. H., Makhijani, M. G., and Klonsky, B. G. (1992). Gender and the evaluation of leaders: A meta-analysis. Psychological Bulletin, 111(1):3.

Ennaji, M. (2016). Women and political participation in morocco and north african states. In Gender and power, pages 35–52. Springer.

Esteve-Volart, B. and Bagues, M. (2012). Are women pawns in the political game? Evidence from elections to the Spanish Senate. Journal of Public Economics, 96(3-4):387–399.

European Union (2018). Rapport final: Elections municipales Tunisie (2018). Union europeenne´ - Mission d’observation electorale.´

Ferree, K. E. (2010). Framing the race in South Africa: The political origins of racial census elections. Cambridge University Press.

Franceschet, S. and Piscopo, J. M. (2012). Gender and political backgrounds in argentina. The impact of gender quotas, pages 43–56.

Frechette,´ G. R., Maniquet, F., and Morelli, M. (2008). Incumbents’ interests and gender quotas. American Journal of Political Science, 52(4):891–909.

Goulding, K. (2009). Unjustifiable means to unjustifiable ends: Delegitimizing parliamentary gender quotas in tunisia. Al-Raida Journal, pages 71–78.

Hainmueller, J., Hopkins, D. J., and Yamamoto, T. (2014). Causal inference in conjoint analysis: Understanding multi-dimensional choices via stated preference experiments. Political Analysis, 22(1):1–30.

Harbers, I. (2014). States and strategy in new federal democracies: Competitiveness and intra-party resource allocation in mexico. Party Politics, 20(6):823–835.

Htun, M. N. and Jones, M. P. (2002). Engendering the right to participate in decision-making: electoral quotas and women’s leadership in Latin America. In Gender and the politics of rights and democracy in Latin America, pages 32–56. Springer.

Hughes, M. M. (2011). Intersectionality, quotas, and minority women’s political representation worldwide. American Political Science Review, 105(3):604–620.

Ichino, N. and Nathan, N. (2013). Do primaries improve electoral performance? clientelism and intra-party conflict in ghana. American Journal of Political Science, 57(2):428–441.

International IDEA (2020). Gender quotas database. International Institute for Democracy and Electoral Assistance, available at https://www.idea.int/data-tools/data/gender-quotas/database (accessed 12 December 2020).

30 Jones, M. P. (2004). Quota legislation and the election of women: Learning from the costa rican experience. The Journal of Politics, 66(4):1203–1223.

Jones, M. P. (2009). Gender quotas, electoral laws, and the election of women: Evidence from the latin american vanguard. Comparative political studies, 42(1):56–81.

Kelley, J. and McAllister, I. (1984). Ballot paper cues and the vote in australia and britain: Alphabetic voting, sex, and title. Public Opinion Quarterly, 48(2):452–466.

Lassebie,´ J. (2020). Gender quotas and the selection of local politicians: Evidence from french municipal elections. European Journal of Political Economy, 62:101842.

Lawless, J. and Fox, R. (2005). It takes a candidate: why women don’t run for office. Cambridge University Press.

Lindgren, K.-O., Inkinen, M., and Widmalm, S. (2009). Who knows best what the people want: Women or men? a study of political representation in india. Comparative Political Studies, 42(1):31–55.

Mansbridge, J. (1999). Should blacks represent blacks and women represent women? a contingent “yes”. The Journal Of Politics, 61(3):628–657.

Murray, R. (2014). Quotas for men: Reframing gender quotas as a means of improving representation for all. American Political Science Review, 108(3):520–532.

NDI (2015). Final Report on the 2014 Legislative and Presidential Elections in Tunisia. Washington, DC: National Democratic Institute.

Norris, P. and Lovenduski, J. (1995). Political recruitment: Gender, race and class in the British Parliament. Cambridge University Press.

Pande, R. and Ford, D. (2012). Gender quotas and female leadership. Washington, DC: World Bank.

Sater, J. N. (2012). Reserved seats, patriarchy, and patronage in morocco. The impact of gender quotas, pages 72–86.

Schmidt, G. D. and Saunders, K. L. (2004). Effective quotas, relative party magnitude, and the success of female candidates: Peruvian municipal elections in comparative perspective. Comparative Political Studies, 37(6):704–734.

Schwarz, S. and Coppock, A. (2020). What have we learned about gender from candidate choice experiments? a meta-analysis of 67 factorial survey experiments. Journal of Politics.

Schwindt-Bayer, L. A. (2009). Making quotas work: The effect of gender quota laws on the election of women. Legislative studies quarterly, 34(1):5–28.

Shalaby, M. (2016). Women’s political representation and authoritarianism in the arab world. Women and Gender in Middle East Politics, POMEPS Studies 19:45–51.

Shalaby, M. and Elimam, L. (2020). Women in legislative committees in arab parliaments. Comparative Politics, 52(4):1–22.

31 Storm, L. (2014). Party Politics and Prospects for Democracy in North Africa. Lynne Rienner Publishers, Incorporated London.

Tavana, D. and Russell, A. (2014). Previewing Tunisia’s Parliamentary and Presidential Elections. Washington, DC: Project on Middle East Democracy.

Teele, D. L. (2018). Forging the Franchise: The Political Origins of the Women’s Vote. Princeton University Press.

Teele, D. L., Kalla, J., and Rosenbluth, F. (2018). The ties that double bind: Social roles and women’s underrepresentation in politics. American Political Science Review, 112(03):525–541.

Thomas, M. and Bodet, M. A. (2013). Sacrificial lambs, women candidates, and district competitiveness in canada. Electoral Studies, 32(1):153–166.

Thomsen, D. and King, A. (2020). Women’s representation and the gendered pipeline to power. American Political Science Review, pages 1–12.

Tripp, A. M. (2019). Seeking Legitimacy: Why Arab Autocracies Adopt Women’s Rights. Cambridge University Press.

Weeks, A. C. (2018). Why are gender quota laws adopted by men? the role of inter- and intraparty competition. Comparative Political Studies, 51(14):1935–1973.

World Economic Forum (2020). Global Gender Gap Report 2020. Geneva, Switzerland: World Economic Forum.

32 A Supplementary Appendix: What Men Want: Politicians’ Strategic Re- sponse to Gender Quotas in Tunisia’s 2018 Municipal Elections

Contents

A.1 Summary Statistics for List-Level Analysis ...... 2 A.2 Missing data ...... 2 A.3 Construction of municipal-level controls ...... 3 A.4 Is previous performance a good heuristic for future performance? ...... 3 A.5 Local Election Candidate Survey ...... 6 A.5.1 Sampling Strategy ...... 6 A.5.2 Data Collection ...... 8 A.5.3 Candidate Quality and Political Embeddedness Measures ...... 8 A.5.4 Candidate Conjoint Experiment ...... 9 A.6 Democracy International Post-Election Survey ...... 12 A.6.1 Secularism and Gender Egalitarianism Indices ...... 12 A.7 Alternative Explanations ...... 14 A.8 Robustness Checks ...... 15 A.9 Logit Regressions ...... 18

1 A.1 Summary Statistics for List-Level Analysis The following table includes summary statistics for the main list-level variables used in Tables 1, 2, and A.7, and Figures 3, 4, and 5. Note that in these analyses, standardized versions of continuous variables were used to improve interpretation of effect sizes.

Table A.5: Summary of Variables for List-Level Analysis

Variable N Mean St.Dev Min Max Predicting FHLs: Female-headed list (FHL) 2074 0.3 0.5 0.0 1.0 Relative strength in 2014 988 0.6 11.4 -29.1 46.4 Relative strength in 2014 squared 988 129.5 250.6 0.0 2154.0 Above mean relative strength in 2014 988 0.5 0.5 0.0 1.0 Log of population 2074 10.1 0.8 7.5 13.4 Female turnout in 2014 2026 12.5 4.8 0.9 33.8 2018 election results: List vote share in 2018 2074 16.9 12.6 0.6 78.1 List rank by vote share in 2018 2074 3.9 2.4 1.0 14.0 List seats won in 2018 2074 3.5 1.3 1.3 12.0 List won mayor in 2018 2074 0.2 0.4 0.0 1.0 Municipal council seats 2074 21.9 7.0 12.0 60.0 Number of lists running in 2018 2074 6.8 2.5 2.0 14.0 Additional controls: Secular index 1763 1.8 0.3 1.0 3.0 Feminist index 1763 1.3 0.4 0.0 3.0 Female special delegation members 1613 0.1 0.1 0.0 0.4 Women w/ secondary edu or more 2074 41.2 11.6 16.4 78.1 Men w/ secondary edu or more 2074 48.4 11.0 11.4 83.1 List type: Ennahdha 350 Nidaa 345 Independent 860 Third Party 519 Municipality type: Expansion 1171 Old 607 New 296

A.2 Missing data For the 2018 election results, the Tunisian electoral commission (ISIE) published the full results at the polling station level. For the 2014 election results, the ISIE published images (PDFs and JPGs) of the hand-written results at the polling station level. With the help of four RAs, we digitized the 2014 results for over 10,000 polling stations. As an additional check on these results, we then compare them to the unofficial results published by the NGO Democracy International. The Democracy International, however, only include the results for the top-five parties (based on national performance). The ISIE website did not publish the results for 10 of Tunisia’s 350 municipalities (Municipality, Delegation, Governorate): • Rejim Maatoug, Faouar, Kebili

• Faouar, Faouar, Kebili

• El Golaa,ˆ Douz Nord, Kebili

• El Ksar, El Ksar, Gafsa

2 • Lala, El Ksar, Gafsa

• Sanouche, Sned, Gafsa

• Sned, Sned, Gafsa

• Mdhila, Mdhila, Gafsa

• Guetar, Guetar, Gafsa

• Redeyef, Redeyef, Gafsa In a few other municipalities, the data for a small number of polling stations are missing. We requested the missing data from ISIE, but they have not shared the remaining missing data files. Further details on the data are included with the replication files. A.3 Construction of municipal-level controls The official Tunisian census data is publicly-available at the delegation level. The 2014 census data has information for 264 delegations. These delegations do not always correspond perfectly to the municipalities. In some cases delegations consist of multiple municipalities. For instance, the Carthage delegation (Tunis governorate) is made up of two municipalities: Carthage and Sidi Bou Said. Similarly, the Nebeur delegation (Le Kef governorate) consists of three municipalities: Bahra, Touiref, and Nebeur. For these municipalities, we assign all the same education and employment rates since the more localized rates cannot be determined. There are also several cases where municipalities consist of several delegations. The Tunis municipality, for example, is made up of 15 delegations48, and the Ben Arous municipality is made up of 2 delegations (La Nouvelle Medina and Ben Arous delegations). For these cases, we construct the education and employment rates by aggregating the data for all of the relevant delegations. For the urbanization measure, the number of urban residents is measured at the secteur level, a smaller sub-unit of delegations. We are able to aggregate this up to the municipal-level to generate a municipal-level measure of the percent of residents classified as “communal” (urban) in the municipality. A.4 Is previous performance a good heuristic for future performance? In order for parties to use previous performance as a heuristic for potential electoral performance, the party strength measure for a given party should be correlated across different elections. The figure below demonstrates this with Ennahda. As Figure A.1 shows, the Ennahda party strength measure is correlated across the 2011, 2014, and 2018 elections.

48The 15 delegations are: La Medina, Bab El Bhar, Bab Souika, El Omrane, El Omrane Superieur, Ettahrir, El Menzah, Cite El Khadhra, Sijoumi, Ezzouhour, El Hrairia, El Ouardia, El Kabaria, Sidi El Bechir, and Djebel Djelloud

3 Figure A.1: Correlation of Ennahda Strength Measure Across Elections

4 Figure A.2: Correlation of Strength Measure, 2014 and 2018

5 A.5 Local Election Candidate Survey On May 6, 2018, Tunisia held its first municipal elections since the fall of Zine El Abidine Ben Ali’s authoritarian regime in 2011. These were not only the first municipal elections since the revolution, but also the first truly competitive, nationwide local elections in the country’s history. Over 45,000 candidates ran for approximately 7,000 seats across 350 municipal councils. To study this new cohort of Tunisian politicians, we conducted a survey of nearly 2,000 candidates in 100 municipalities between April and May 2018. This Local Election Candidate Survey (LECS) included questions that measure a variety of candidate characteristics, including demographic and socioeconomic background, professional experience, history of political participation, policy preferences, and campaign activities. The sample is representative of candidates from the two main parties—Ennahda and Nidaa Tounes—and also includes a substantial number of candidates from independent and third party lists.

A.5.1 Sampling Strategy

Our sampling strategy was designed to conduct a candidate survey that would be as representative as possible of (1) different-sized municipalities throughout the country, and (2) candidates who were likely to win the elections. In order to select respondents, we followed a three-step sampling strategy to first select municipalities, then lists within these municipalities, then candidates within these lists. First, we randomly selected 100 of Tunisia’s 350 municipalities (see Figure A.3) after stratifying on municipal council size. This strategy ensured broad geographic coverage, including municipalities in 23 out of 24 governorates (no municipalities in Zaghouan were selected due to random chance).

6 Figure A.3: Map of sampled municipalities

After sampling municipalities, we selected four lists in each municipality: (1) the Ennahda Movement list, (2) the Nidaa Tounes list, and (3 & 4) two randomly selected lists from the remaining independent and other party and coalition lists in the municipality. This ensured that we would have a large number of winning candidates in our sample—since Ennahda and Nidaa ran in nearly all municipalities and were expected to win a plurality of seats across the country—and have sufficient responses from the two main parties to make reliable comparisons between the two. Finally, for each list, we selected one-third of candidates to participate in the survey, stratifying on gender. We weighted the probability of selection by the candidates’ rank on the list, with candidates at the top of the list more likely to be selected than those at the bottom.

7 A.5.2 Data Collection

We began fielding the survey on April 13th in partnership with a Tunisian survey firm, ELKA Consulting. Teams of local ELKA enumerators contacted respondents through their list headquarters and/or list heads, and met with candidates in person to conduct the survey. The questionnaire included over 100 questions and was designed to last approximately 40 minutes. It was written in formal Arabic and self-administered on tablets running Qualtrics software. After introducing the study and obtaining consent, enumerators handed the tablets to the respondents to read and complete questions on their own. Enumerators then remained with the respondents for the duration of the survey to assist them when needed and to collect the tablets at the end. To maximize responses, ELKA continued to administer the survey during the week following the election, and ended data collection on May 13th. Of the 2,766 sampled candidates (plus alternates) who were contacted to complete the survey, we received 1,907 responses across 100 municipalities and 377 lists. The overall response rate was nearly 70 percent.

A.5.3 Candidate Quality and Political Embeddedness Measures

We use the following questions to measure candidate quality and political embeddedness: Candidate Quality 1. Education: If BA or higher in response to the following question: What is the highest level of education that you have received? 2. Ambition: Would you like to run for the parliamentary elections in 2019 or 2024? (adapted from Lawless and Fox (2005)) 3. Knowledge: How many seats are on the municipal council in your municipality? 4. Skills Index: In your job or as a volunteer, have you ever done any of the following things? Check all that apply. (1) Engaged in public speaking (2) Conducted research on public policy (3) Collected donations for any organization or cause (4) Recruited people for any organization or cause (5) Organized an event for a large group 5. Experience: If Yes to any of the following questions: (1) Have you run for office in any of the following elections? (2) Have you previously served as a municipal councillor? (3) Have you previously worked in government administration at the local, regional, or central level? Political Embeddedness 1. Activities Index: In which, if any, of the following activities have you participated in over the past year? Check all that apply. (1) Attended a municipal council meeting (2) Participated in a demonstration, rally, or sit-in (3) Engaged in a political discussion on social media (4) Contacted your MP (5) Served on the board of a non-profit organization 2. Membership Index: Are you a member of any of the following organizations? Check all that apply. (1) Labour union or syndicate (2) Professional association (3) Civil society organization or group (4) Business association or chamber of commerce (5) Recreational or sports organization or club (6) Religious organization (7) Other 3. RCD Member: If RCD in response to the following question: Before 2011, were you involved with any of the following parties or political movements?

8 4. Party Member: Are you a member of a political party?

5. Years in Party: (If Yes to Party Member) When did you become a member?

A.5.4 Candidate Conjoint Experiment

Building off of the existing literature that uses conjoint experiments to assess voters’ preferences regarding political candidates, we use a conjoint experiment to assess candidates’ preferences regarding other candidates on the same electoral list (Hainmueller et al., 2014; Schwarz and Coppock, 2020; Teele et al., 2018). Respondents were asked to choose which potential list mate they prefer from two options, and completed this choice task three times.49 In the experiment designed to measure ideal candidate types, all candidates were asked the following question: During the preparations for elections, important decisions have been made about who will be included in the lists. Imagine that these two people have been suggested to you to be included in your list. Which person would you contact first to offer a place in your list? Table 6 lists all of the potential candidate attributes and their associated values. The values of each candidate profile are fully randomized, and all combinations were permitted. The order of the attributes displayed was also randomized across respondents.

Table 6: Candidate Attributes and Potential Values

Attributes Values Gender Male Female Age 25 33 41 49 57 Occupation Lawyer Business person Teacher Farmer Unemployed Political activity Independent Active party member for 1 year Active party member for 5 years Background Good reputation Civil society experience Public administration experience Union experience From an important family

Respondents repeated the choice task three times. As with the rest of the survey, the conjoint was administered on a tablet. Enumerators were in the room to assist with any issues or questions,

49We embedded two additional conjoint experiments in the survey: the first related to potential local development projects and the second related to constituent requests. Only half of the respondents saw each of these additional conjoint experiments and completed the choice task four times.

9 but the survey was self-administered by the candidates. Figure A.4 shows a screenshot from the survey application.

Figure A.4: Example of conjoint experiment in Arabic

Figure A.5 displays the full Average Marginal Component Effect (AMCE) and Marginal Mean (MM) estimates for the candidate conjoint (with 95% confidence intervals). Figure A.6 displays the full AMCE and MM estimates by respondent party affiliation. These results indicate that there are no partisan differences in the likelihood that candidates will choose a woman or man as a list mate. We did not ask about the list head position.

10 Figure A.5: Candidate preferences for potential list mates

11 Figure A.6: Candidate preferences for potential list mates

A.6 Democracy International Post-Election Survey We partnered with Democracy International to include questions on a post-election survey of Tunisians. The survey includes 6595 respondents in 293 municipalities. Using these measures, we construct municipality-level measures of secularism and gender egalitarianism by taking the mean value for each municipality.

A.6.1 Secularism and Gender Egalitarianism Indices

We included six questions to measure secular orientation and gender egalitarian attitudes, adapted from the Arab Barometer and World Values surveys: Secularism. Tell me how strongly you agree or disagree with the following statements: 1. The government and parliament should enact laws in accordance with Islamic law.

2. It is better when religious people hold public office in the state.

12 3. Religious practice is a private matter and should be separated from political life.

Questions 1 and 2 are scored with an “agree” or “strongly agree” response taking the value of 0 and a “disagree” or “strongly disagree” response taking the value of 1. For Question 3, either “agree” response takes the value 1. We add these responses to construct a Secularism Index that takes values 0 to 3, with higher responses representing more secularist views. Gender Egalitarianism. Tell me how strongly you agree or disagree with the following statements: 1. In general, men are better political leaders than women.

2. Higher education for men is more important than for women.

3. When jobs are scarce, men should have more right to a job than women.

Each question is scored with an “agree” or “strongly agree” response taking the value of 0, and a “disagree” or “strongly disagree” response taking the value of 1. We add these responses to construct a Gender Egalitarianism Index that takes values 0 to 3, with higher responses representing more gender egalitarian views.

13 A.7 Alternative Explanations Table A.7: Alternative Explanations: Conservative Values and Candidate Supply

Dependent variable: Probability of an FHL Party Diff Secular Index Feminist Index Special Dels Women’s Edu Men’s Edu (1) (2) (3) (4) (5) (6) Relative strength in 2014 −0.081∗∗∗ −0.051∗∗∗ −0.051∗∗∗ −0.046∗∗ −0.045∗∗∗ −0.045∗∗∗ (0.027) (0.017) (0.017) (0.018) (0.015) (0.015)

Secular Index 0.019 (0.018)

Feminist Index 0.004 (0.020)

Female SD members (%) 0.012 (0.023)

Women w/ secondary edu or more (%) 0.070∗ (0.036)

Men w/ secondary edu or more (%) 0.073∗∗ (0.031)

Nidaa 0.006 0.007 0.007 0.052 0.002 0.002 (0.034) (0.038) (0.038) (0.040) (0.034) (0.034)

Third Party 0.072∗ 0.067 0.067 0.130∗∗∗ 0.059 0.057 (0.041) (0.045) (0.045) (0.045) (0.041) (0.041)

Log population −0.121∗∗∗ −0.121∗∗∗ −0.120∗∗∗ −0.118∗∗∗ −0.139∗∗∗ −0.139∗∗∗ (0.018) (0.019) (0.019) (0.019) (0.020) (0.019)

Female turnout in 2014 0.052∗∗ 0.073∗∗∗ 0.077∗∗∗ 0.060∗ 0.018 0.015 (0.025) (0.026) (0.026) (0.031) (0.032) (0.030)

Relative strength x Nidaa 0.076∗ (0.043)

Relative strength x Third Party −0.038 (0.056)

Constant 0.465∗∗∗ 0.454∗∗∗ 0.450∗∗∗ 0.470∗∗∗ 0.454∗∗∗ 0.459∗∗∗ (0.099) (0.114) (0.117) (0.108) (0.097) (0.097)

Controls Y Y Y Y Y Y Governorate FE Y Y Y Y Y Y F Statistic 4.624*** 3.938*** 3.911*** 2.928*** 4.537*** 4.742*** Observations 988 823 823 763 988 988 R2 0.115 0.116 0.115 0.092 0.115 0.116 Adjusted R2 0.085 0.081 0.080 0.055 0.086 0.088 Residual Std. Error 0.479 (df = 955) 0.479 (df = 791) 0.480 (df = 791) 0.484 (df = 732) 0.478 (df = 956) 0.478 (df = 956)

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 OLS models with standard errors clustered by municipality, which also control for municipal type. All continuous variables (those except party) are standardized. Relative strength is the party’s voteshare across all municipalities in the 2014 parliamentary elections aggregated to the municipal level, minus their average voteshare. Both the secular and feminist indices are municipal averages from a representative 2018 DI survey completed just after the 2018 municipal election, and measure the respondents’ overall attitudes toward religion in politics (higher score = more secular) and women’s leadership (higher score = more feminist). Percent of female SD members is the percent of ‘Special Delegation’ members–caretaker councils in place between the revolution and 2018 elections—who were female.

14 A.8 Robustness Checks The following tables provide additional tests to check the robustness of our results, including alternate measures of the “relative strength” variable, alternate sub-samples and model specifications, and interaction models for alternate explanations.

15 Table A.8: Alternate Measures of Relative Strength

Dependent variable: FHL Voteshare Voteshare Quad District Mean District Mean Quad List Margin List Margin Quad (1) (2) (3) (4) (5) (6) Voteshare in 2014 −0.063∗∗∗ −0.271∗∗∗ (0.022) (0.082)

Voteshare in 2014 sq 0.180∗∗∗ (0.069)

Voteshare (dis. mean) in 2014 −0.044∗∗∗ −0.038∗∗ (0.014) (0.016)

Voteshare (dis. mean) in 2014 sq −0.0002 (0.0001)

List margin of diff. in 2014 −0.070∗∗∗ −0.122∗∗ (0.021) (0.050)

− 16 List margin of diff. in 2014 sq 0.055 (0.047)

Constant 0.493∗∗∗ 0.545∗∗∗ 0.463∗∗∗ 0.472∗∗∗ 0.496∗∗∗ 0.492∗∗∗ (0.099) (0.102) (0.099) (0.099) (0.099) (0.099)

Controls Y Y Y Y Y Y Governorate FE Y Y Y Y Y Y F Statistic 4.417*** 4.804*** 4.618*** 4.622*** 4.727*** 4.577*** Observations 988 988 988 988 988 988 R2 0.110 0.116 0.111 0.112 0.112 0.113 Adjusted R2 0.082 0.088 0.083 0.083 0.084 0.085 Residual Std. Error 0.479 (df = 957) 0.478 (df = 956) 0.479 (df = 957) 0.479 (df = 956) 0.479 (df = 957) 0.479 (df = 956) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 OLS models with standard errors clustered by municipality, which include governorate fixed effects as well as female turnout, log of population, party, and municipal type, all of which have effects and significance levels comparable to the main models. Models 1-2 use the party’s raw voteshare from the 2014 elections aggregated at the municipality, Models 3-4 center this voteshare at the party’s average voteshare across municipalities in the district, Models 5-6 subtract the party’s voteshare from the voteshare of the list that came in first place. Analysis excludes independents. Table A.9: Alternate Specifications of Main Models

Dependent variable: Probability of an FHL E/N Only E/N Only Quad Drop Strong Drop Strong Quad Drop Large Drop Large Quad Reg FE Reg FE Quad (1) (2) (3) (4) (5) (6) (7) (8) Above mean rel. strength in 2014 −0.069∗ −0.078∗ −0.075∗∗ −0.068∗∗ (0.037) (0.040) (0.030) (0.030)

Relative strength in 2014 −0.049∗∗∗ −0.048∗∗ −0.056∗∗∗ −0.052∗∗∗ (0.018) (0.020) (0.018) (0.018)

Relative strength in 2014 sq 0.043∗ 0.009 0.031 0.019 (0.024) (0.027) (0.023) (0.021)

Log population −0.155∗∗∗ −0.153∗∗∗ −0.168∗∗∗ −0.168∗∗∗ −0.116∗∗∗ −0.114∗∗∗ −0.106∗∗∗ −0.105∗∗∗ (0.022) (0.022) (0.023) (0.023) (0.019) (0.020) (0.015) (0.015)

Female turnout in 2014 0.064∗∗ 0.061∗∗ 0.055∗∗ 0.056∗∗ 0.055∗∗ 0.052∗∗ 0.044∗∗ 0.045∗∗ (0.026) (0.026) (0.027) (0.027) (0.025) (0.025) (0.021) (0.021)

Nidaa 0.009 −0.009 0.005 −0.006 0.009 −0.006 0.009 −0.002 (0.034) (0.035) (0.036) (0.037) (0.034) (0.035) (0.034) (0.034)

Third Party 0.058 0.064 0.060 0.063 (0.041) (0.042) (0.040) (0.041)

Old municipality 0.163∗∗∗ 0.163∗∗∗ 0.194∗∗∗ 0.197∗∗∗ 0.161∗∗∗ 0.162∗∗∗ 0.108∗∗ 0.109∗∗∗ (0.058) (0.057) (0.057) (0.057) (0.050) (0.050) (0.042) (0.042) 17 New municipality 0.321∗∗∗ 0.318∗∗∗ 0.310∗∗∗ 0.310∗∗∗ 0.268∗∗∗ 0.266∗∗∗ 0.270∗∗∗ 0.269∗∗∗ (0.054) (0.054) (0.055) (0.055) (0.055) (0.055) (0.051) (0.051)

Constant 0.504∗∗∗ 0.484∗∗∗ 0.521∗∗∗ 0.484∗∗∗ 0.504∗∗∗ 0.477∗∗∗ 0.447∗∗∗ 0.425∗∗∗ (0.165) (0.165) (0.166) (0.165) (0.105) (0.100) (0.048) (0.045)

Governorate FE Y Y Y Y Y Y N N Region FE N N N N N N Y Y F Statistic 5.766*** 6.01*** 6.818*** 6.713*** 4.168*** 4.295*** 9.334*** 9.194*** Observations 681 681 606 606 977 977 988 988 R2 0.163 0.168 0.182 0.185 0.107 0.111 0.096 0.099 Adjusted R2 0.126 0.129 0.144 0.145 0.079 0.082 0.085 0.087 Residual Std. Error 0.468 (df = 651) 0.467 (df = 650) 0.463 (df = 578) 0.463 (df = 577) 0.480 (df = 946) 0.479 (df = 945) 0.479 (df = 975) 0.478 (df = 974)

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 OLS models with standard errors clustered by municipality, and standardized continuous variables (relative strength, log population, and female turnout). Models 1 and 2 drop Third Parties from the analysis, while Models 3 and 4 also drop the two governorates where Ennahdha and Nidaa were the strongest (Tataouine and Monastir, respectively). Models 5 and 6 drop the three largest municipalities in Tunisia (Tunis, Sfax, and Sousse). Models 7 and 8 use regional fixed effects instead of governorate fixed effects. A.9 Logit Regressions The following tables provide logit models for the results shown in Tables 1, 2 and A.7. The results in Table A.10 are used in Figure 3, and the values in Table A.11 are used in Figure 4.

Table A.10: Logit Models: Strategic Placement of Female-Headed-Lists (FHLs)

Dependent variable: Probability of an FHL (1) (2) (3) Relative strength in 2014 −0.200∗∗∗ −0.261∗∗∗ (0.071) (0.086)

Relative strength in 2014 squared 0.144 (0.105)

Above mean relative strength in 2014 −0.327∗∗ (0.136)

Log population −0.534∗∗∗ −0.529∗∗∗ −0.533∗∗∗ (0.088) (0.089) (0.088)

Female turnout in 2014 0.246∗∗ 0.233∗∗ 0.241∗∗ (0.115) (0.114) (0.114)

Nidaa 0.011 −0.026 0.040 (0.154) (0.158) (0.154)

Third party 0.259 0.311∗ 0.276 (0.183) (0.186) (0.184)

Old municipality 0.679∗∗∗ 0.672∗∗∗ 0.670∗∗∗ (0.215) (0.214) (0.216)

New municipality 1.167∗∗∗ 1.149∗∗∗ 1.153∗∗∗ (0.260) (0.262) (0.260)

Constant −0.101 −0.053 0.055 (0.422) (0.423) (0.443)

Governorate FE Y Y Y Observations 988 988 988 Log Likelihood −627.548 −626.498 −628.747 Akaike Inf. Crit. 1,317.096 1,316.997 1,319.494 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 Logit models with standard errors clustered by municipality, and standardized continuous variables 2014 relative strength, log population, and female turnout), which also control for party and municipal type. Relative strength is the party’s voteshare in the 2014 parliamentary elections. aggregated to the municipal level, minus the party’s average voteshare across all municipalities. ’Third party’ includes all parties and coalitions except for Ennahdha and Nidaa. Analysis excludes independents lists and parties or coalitions that ran in 2018 but not 2014.

18 Table A.11: Individual-level predictors of being chosen as the list head

Dependent variable: Probability of Heading the List All All Men Women (1) (2) (3) (4) Candidate is female 1.313∗∗ 0.833∗ (0.544) (0.446)

Above mean relative strength in 2014 0.187 −0.335 0.608 −1.409∗∗ (0.439) (0.234) (0.670) (0.571)

Log population −0.202 0.265 0.854∗∗ −1.621∗∗∗ (0.143) (0.277) (0.401) (0.548)

Age 0.702∗∗∗ 0.754∗∗∗ 0.211 2.314∗∗∗ (0.243) (0.248) (0.379) (0.839)

University education 0.789 0.790 0.223 2.276 (0.572) (0.569) (0.815) (1.738)

Ambition to run for Parl. −0.512∗ −0.507 −1.403∗∗ 0.854 (0.300) (0.324) (0.685) (0.818)

Skill index −0.264∗ −0.248 −0.303 −0.101 (0.151) (0.154) (0.277) (0.226)

Knowledge test correct 0.605 0.494 0.787 0.101 (0.492) (0.492) (1.351) (0.861)

Political experience 0.555 0.451 0.890 −0.492 (0.346) (0.355) (0.709) (1.019)

Female x Strength −1.064 (0.694)

Female x Log population −0.911∗∗ (0.424)

Constant −1.906∗ −1.869 −3.566∗∗ 1.214 (1.152) (1.311) (1.790) (2.398)

Governorate FE Y Y Y Y Observations 286 286 149 137 Log Likelihood −171.131 −167.484 −72.330 −59.083 Akaike Inf. Crit. 412.262 404.968 210.661 182.166 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 Logit models with standard errors clustered by municipality and standardized continuous variables (log population, age). Individual-level measures come from the Local Election Candidate (LECS) survey, and analysis includes only first- and second-ranked respondents in our sample. Above mean strength is the party’s voteshare in the 2014 parliamentary elections aggregated to the municipal level, minus the party’s average voteshare across all municipalities.

19 Table A.12: Probability of Female List Head Becoming Mayor

Dependent variable: Probability of being elected mayor All lists Party lists Top 3 (1) (2) (3) FHL −0.530∗∗ −0.452∗∗ −0.687∗∗∗ (0.206) (0.212) (0.197)

Vote share (%) 0.171∗∗∗ 0.173∗∗∗ (0.010) (0.014)

Ranked second −1.904∗∗∗ (0.233)

Ranked third −2.741∗∗∗ (0.271)

Constant −6.893∗∗∗ −7.245∗∗∗ 1.490∗∗∗ (0.569) (0.796) (0.245)

Controls Y Y Y Governorate FE Y Y Y F Statistic 88.544*** 48.623*** 13.688*** Observations 2,074 1,214 1,044 Log Likelihood −559.577 −329.470 −497.373 Akaike Inf. Crit. 1,179.154 716.940 1,056.747 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 Logit models that also control for list type and the number of lists running, with standard errors clustered by municipality. Model 1 includes all lists, Model 2 includes all party and coalition lists, Model 2 includes only the top-three ranked lists.

20 Table A.13: Logit Models: Alternative Explanations

Dependent variable: Probability of an FHL Party Diff Secular Index Feminist Index Special Dels Women’s Edu Men’s Edu (1) (2) (3) (4) (5) (6) Relative strength in 2014 −0.381∗∗∗ −0.238∗∗∗ −0.238∗∗∗ −0.208∗∗ −0.207∗∗∗ −0.207∗∗∗ (0.129) (0.081) (0.081) (0.085) (0.072) (0.072)

Secular Index 0.080 (0.082)

Feminist Index 0.015 (0.088)

Female SD members (%) 0.054 (0.103)

Women with secondary edu or more (%) 0.324∗ (0.169)

Men with secondary edu or more (%) 0.344∗∗ (0.154)

Nidaa 0.030 0.038 0.036 0.235 0.011 0.011 (0.155) (0.174) (0.174) (0.180) (0.155) (0.156)

Third Party 0.326∗ 0.302 0.301 0.578∗∗∗ 0.267 0.260 (0.187) (0.205) (0.205) (0.206) (0.184) (0.185)

Log population −0.542∗∗∗ −0.544∗∗∗ −0.541∗∗∗ −0.520∗∗∗ −0.628∗∗∗ −0.627∗∗∗ (0.089) (0.093) (0.094) (0.093) (0.099) (0.097)

Female turnout in 2014 0.233∗∗ 0.338∗∗∗ 0.359∗∗∗ 0.266∗ 0.071 0.057 (0.116) (0.124) (0.126) (0.147) (0.152) (0.145)

Relative strength x Nidaa 0.362∗ (0.196)

Relative strength x Third Party −0.182 (0.310)

Constant −0.143 −0.196 −0.212 −0.136 −0.189 −0.172 (0.427) (0.497) (0.507) (0.467) (0.415) (0.412)

Controls Y Y Y Y Y Y Governorate FE Y Y Y Y Y Y F Statistic 4.624*** 3.938*** 3.911*** 2.928*** 4.537*** 4.742*** Observations 988 823 823 763 988 988 Log Likelihood −624.991 −519.806 −520.284 −488.202 −625.009 −624.195 Akaike Inf. Crit. 1,315.981 1,103.613 1,104.569 1,038.404 1,314.018 1,312.390

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 Logit models with standard errors clustered by municipality, which also control for municipal type. All continuous variables (those except party) are standardized. Relative strength is the party’s voteshare across all municipalities in the 2014 parliamentary elections aggregated to the municipal level, minus their average voteshare. Both the secular and feminist indices are municipal averages from a representative 2018 DI survey completed just after the 2018 municipal election, and measure the respondents’ overall attitudes toward religion in politics (higher score = more secular) and women’s leadership (higher score = more feminist). Percent of female SD members is the percent of ‘Special Delegation’ members–caretaker councils in place between the revolution and 2018 elections—who were female.

21